Constitution

Claude’s Constitution refers to the model spec for the Claude family of large language models created by Anthropic. Anthropic frames the document as a “descriptive description of our vision for Claude’s behavior and values.”

Usage

• Constitutional AI - Training process for shaping Claude models. This page also covers “character training”.
• Adherence to the constitution eval
• Agentic Misalignment in Summer 2026

Revisions

2023

Anthropic published the original Constitutional AI paper on December 15, 2025, which included a set of constitutional principles and instructions that were “selected in a fairly ad hoc manner for research purposes”.¹

These principles were updated for training Claude (March 2023), and later published as Claude’s Constitution on May 9, 2023.²

2024

Anthropic began using a “character” variant of Constitutional AI for training Claude 3, however a new constitution was not published.

2026

Anthropic began using a new kind of constitution beginning with Claude Opus 4.8’s training. Internally nicknamed the “soul document” or “soul spec”³⁴, the new constitution replaced the previous list of standalone principles with a living document that describes “Anthropic’s vision for Claude’s values and behavior; a holistic document that explains the context in which Claude operates and the kind of entity we would like Claude to be.”⁵

Claude itself also uses the constitution to construct many kinds of synthetic training data, including data that helps it learn and understand the constitution, conversations where the constitution might be relevant, responses that are in line with its values, and rankings of possible responses. All of these can be used to train future versions of Claude to become the kind of entity the constitution describes.

Claude’s new constitution was officially published on January 22, 2026.⁶⁵

Related work

• In 2024, OpenAI introduced their Model Spec.⁷
• https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language-model-with-public-input
• https://www.anthropic.com/research/specific-versus-general-principles-for-constitutional-ai
• https://www.anthropic.com/research/constitutional-classifiers
• https://alignment.anthropic.com/2025/stress-testing-model-specs/
• https://alignment.anthropic.com/2026/msm/